CAPACITY GAINS OF SINGLE ANTENNA INTERFERENCE CANCELLATION IN GSM Martti Moisio 1, Sami Nikkarinen 2 1 2
Nokia Research Center, P.O.Box 407, FIN-00045 NOKIA GROUP, Finland,
[email protected] Nokia Research Center, P.O.Box 407, FIN-00045 NOKIA GROUP, Finland,
[email protected]
Abstract – Single Antenna Interference Cancellation (SAIC) is a very interesting and promising technology currently being standardized in 3GPP as part of GERAN Release 6. This interference reduction technique takes advantage of the latest developments in signal processing and receiver algorithms in a cost-efficient way. In most advantageous interference conditions, SAIC can improve the signal-to-noise ratio over 10 dB. In this paper, we investigate the effect SAIC on the capacity of a typical GSM network. Also, an overview of SAIC algorithms is presented and a new interface for mapping the link and network simulator is described. The network level performance of SAIC is evaluated from both the quality and capacity point of view by using a sophisticated burstlevel dynamic network level simulator. The simulations are performed in a synchronized frequency hopping macro cell network. The simulations results verify that the high link level gains can be turned into significant network capacity gains.
Keywords – GSM, simulation, interference cancellation, SAIC, capacity. I. INTRODUCTION GSM may be seen by someone as an old and mature technology that will soon be surpassed by more novel technologies like Wideband CDMA and CDMA2000 evolutions. However, this is not at all the case – GSM/EDGE is under constant development in the 3GPP standardization bodies. With the introduction of such new features like EDGE and AMR, the capacity and performance of GSM/EDGE Radio Access Network (GERAN) will be at totally different level than what was achievable in e.g. GSM Release 97. Currently 3GPP is working with GSM/EDGE Release 6 that will once again bring many new performance-enhancing features into GERAN networks. One of the most interesting work items in Release 6 is SAIC [1] (Single Antenna Interference Cancellation). A more generic name ARP (Advanced Receiver Performance) is also being used. An
ARP receiver is any receiver that substantially exceeds the conventional receiver performance. In this paper we focus on the effect of SAIC in the network level. In Section II, SAIC is briefly introduced. Section III describes the SAIC network level effects in general. Section IV provides details on the simulation methodology and assumptions. In Section IV, the results are presented and analyzed, and finally in Section V, conclusions are made. II. SAIC ALGORITHM OVERVIEW SAIC is a general term for receiver algorithms that employ interference suppression methods with single antenna in order to improve the receiver performance in presence of cochannel interference. Although nothing prevents to use SAIC in base station, it is normally seen as a method to be used only in Mobile Station (MS). Throughout this paper, we are focusing on SAIC in MS. In the early days of GSM, the networks were mainly coverage limited. Nowadays, due to the huge increase in GSM traffic, networks are becoming more and more interference limited. That is, the dominant factor limiting the capacity of a GSM network is co-channel interference and not (thermal) noise. Hence, in order to maximize the spectral efficiency of a GSM system, any method that reduces the co-channel interference is of vast importance. The actual implementation of SAIC is up to the MS vendor. Commonly known methods used in GSM are for example Joint Detection/Demodulation (JD) and Blind Interference Cancellation (BIC). The basic idea in JD [2] is the simultaneous detection of both the wanted signal and the interfering signal. This is achieved by performing joint channel estimation from the different Training Sequence Codes (TSC) of the signals. Although rather complex, JD is very effective, especially in the presence of one dominant interferer. BIC methods (see e.g. [3]) do not rely on the channel estimation of both signals. Only the wanted signal is processed (i.e. BIC is “blind” for the interfering signal) and the gain is achieved by suppressing the interference from the
wanted signal. Several different BIC methods exist; often they rely on the known waveform of the GSM signal and separation of the signal into real and imaginary parts. BIC method is currently seen as a viable low-complexity option to be used in MS receiver.
Since SAIC receiver is in MS, only the downlink direction is studied.
A proprietary BIC method is assumed throughout in this study. The Release 6 Work Item on SAIC/ARP will focus mainly on GMSK signals. However, SAIC receiver performance in the presence of 8PSK signals should not of course be worse than with conventional receiver. Interference cancellation for 8-PSK-modulated signals is more difficult than with GMSK and it is anticipated that the gains are also smaller.
III. SAIC NETWORK LEVEL EFFECTS When MS has a SAIC receiver it can tolerate lower Carrierto-Interference-Ratio (CIR) than a legacy MS with conventional receiver. This results into a lower Bit Error Rate (BER). In case of speech service, the Frame Error Rate (FER) will get smaller and user perceived quality improves. Signalling channel quality is also improved and, as a result, dropping probability gets smaller. For data services such as GPRS, SAIC naturally improves the user throughput [6]. An interesting phenomenon is that SAIC mobiles in the network can also improve the performance of non-SAIC mobiles [7]. This happens when the mobiles are using quality-based power control. The SAIC mobiles are able to use lower power levels than legacy mobiles, which decrease the interference level in the network. It should be noted that SAIC works best in a synchronized network where the time slots borders between different cells do not overlap. IV. NETWORK SIMULATION METHODOLOGY AND ASSUMPTIONS
Figure 1. The macro cell network deployment scheme. Advanced Radio Resource Management (RRM) schemes are used: random synthesized 1/1 Frequency Hopping (FH) with Mobile Allocation Index Offset (MAIO) management, Discontinuous Transmission (DTX) and Power Control (PC). It is assumed that the network is synchronized. It is important to note that the RRM algorithms in this study are not optimized for SAIC, i.e. the network is not SAIC-aware. Signaling support for MS SAIC/ARP capability is currently being standardized. A more detailed description about mobility, propagation, and shadowing models can be found from [4]. B. Link level mapping tables A very important part of any network simulation is the link level mapping tables. Current state-of-the-art mapping method for conventional receiver is to use 2-stage mapping on burst-level. In the first stage, burst CIR values are mapped onto raw BER values; in the second stage the mean (optionally also the standard deviation) BER (of bursts in the interleaving period) of the radio block is mapped onto Frame Error Probability (FEP):
A. General
BERburst = f (CIRburst )
(1)
In order to quantify the gains of SAIC in the network level, dynamic burst-level network simulations were carried out. The simulation scenario is depicted in Figure 1.
FEPblock = f ( µˆ BER , σˆ BER ) ,
(2)
The network consists of 75 sectorized cells, each cell equipped with 6 TRX (to enable high enough frequency load). Only the non-BCCH layer is simulated and totally 12 frequencies are available for TCH traffic.
where µˆ BER is the mean raw BER and σˆ BER is the standard deviation. In the network simulation, a series of Look-Up-Tables (LUT) are then used based on extensive link level simulations. However, this mapping is not enough for SAIC receiver. It has been shown that although the second stage mapping can
remain the same (coding gain does not change) the simple CIR→BER mapping is not enough for SAIC receiver. Although not perfect assumption in every case, a very good enhancement is to add an extra dimension into the first stage mapping; Dominant-to-rest-Interference-Ratio (DIR). Burstlevel DIR is defined as
DIR =
I max , ∑ I k − I max + ACP × ∑ I l + N 0 k
(3)
l
Table 1 Main simulation parameters. Parameter Carrier frequency
900 MHz
Frequency bandwidth
2.4 MHz
Path loss exponent
3.67
where Imax is the strongest interfering component, ACP is the adjacent channel protection and N0 receiver noise floor. The summations are over k co-channel (fN) and l adjacent channel (fN-1, f N+1) interferers of one burst.
Slow fading standard deviation Slow fading correlation distance
Consequently, the Equation (1) will expand to Equation (4):
Receiver noise floor
BERburst = f (CIRburst , DIRburst )
(4)
An example of this 2-dimensional mapping is shown in Figure 2. For the network simulation LUT, an NxM matrix is constructed where each matrix element contains BER estimation for point (CIRi, DIRj). The middle points can naturally be interpolated.
Value
Cell radius Antenna gain Antenna beamwidth Channel profile ACP (Adjacent Channel Protection)
110 m -110 dBm 750 m 18 dBi 65/90 deg. TU 18 dB
MS speed Average call length
90 sec.
Speech codec
AMR7.4
Voice activity
60%
Power Control Channel allocation
12 hopping frequencies Okumura-Hata model
6 dB
3·108 TDMA frames 3.0 km/h
Simulation Length
Comment
Qualitybased Random
At 3 dBi point Typical Urban 1st adj. taken into account ~23 minutes in real time Exponential distribution. Minimum 5 sec. No link adaptation For DTX Based on RxQual/RxLev
D. Performance criteria Figure 2. Example of 2-dimensional (CIR,DIR)->BER mapping. C. Simulation parameters Main network simulation parameters are presented in Table 1. The simulation setup is typical for GSM/EDGE networks with narrowband deployment (common scenario especially in US markets).
The call was regarded as successful if its average FER (class 1A bits) was less than 2%. Two different FER averaging periods were used: A) whole call averaging, and B) sampled average, where one sample was equal to 4 SACCH periods (1.92 seconds, which is quite commonly used in network monitoring tools). In both cases, the amount of good samples was recorded over all the calls in the whole simulation. The results are reported for points where 95% and 98% of the samples were of good quality.
V. SIMULATION RESULTS This section presents simulation results assuming different SAIC MS penetrations, antenna patterns and Quality of Service (QoS) criteria. In Section A, the ratio of bad quality calls (or samples, in general) is plotted versus network load. Load is presented as Effective Frequency Load (EFL) which is a measure of how much each frequency is loaded on average (for a more detailed description, see e.g. [5]). Section B shows how the performance curves change, if we have a partial SAIC penetration in the network. Finally in Section C, two commonly used operating points are selected in order to calculate the capacity gains for SAIC. These two points are the load levels where 95% and 98% of the users are satisfied (i.e. fulfil the call quality criteria).
Figure 4 Simulation results with 90-degree antenna. (A) = call level averaging, (B) = 1.92s averaging.
A. Call quality versus network load
B. Effect of partial SAIC penetration
Figure 3 and Figure 4 show the call quality versus frequency load for both SAIC and non-SAIC cases when using 65 and 90-degree antennas, respectively. The solid line shows the results for case A (call level averaging) and the dotted line for case B (1.92s averaging).
In order to study the effect of partial SAIC penetration, simulations with 65-degree antenna were rerun with 20%, 40%, 60% and 80% SAIC penetration rates. The results are shown in Figure 5 and they are only for 65-degree antenna.
The results with the wider antenna beamwidth are clearly worse, as expected. The difference between the two quality metrics is surprisingly small, although the shorter averaging seems to always provide better results.
Figure 5. Network quality with 65-degree antenna and partial SAIC penetration. (A) = call level averaging, (B) = 1.92s averaging.
Figure 3. Simulation results with 65-degree antenna. (A) = call level averaging, (B) = 1.92s averaging. It can be also concluded that the FER averaging interval does not have big effect into the simulated SAIC gains.
It can be seen that SAIC gains are almost linearly proportional to the SAIC penetration rate in the network. The earlier findings about the effect of different FER collection periods do not change. C. Capacity gains at selected operating points Figure 6 below shows the SAIC capacity gains with full SAIC penetration. All the combinations are shown: both 65 and 90-degree antennas, 2 different network QoS (95% and 98%) and 2 different FER averaging periods (call level and 1.92 seconds). The gains are higher with tighter network QoS and when wider antenna sectors are used. Although the absolute
capacities are smaller with wider antenna, SAIC gains are higher due to the higher DIR values. In general, the gains are around 50% and almost 60% with 95% and 98% network QoS, respectively. 65 deg. (A)
65 deg. (B)
90 deg. (A)
90 deg. (B)
70% 60%
interference-limited networks. Up-to 50%-60% capacity gains can be achieved in a typical synchronized GSM network. The gains are higher when tighter criteria for user satisfaction are applied. In the simulations, both legacy and SAIC mobiles used the same RRM algorithms and parameters. Higher gains will probably be achievable when RRM algorithms are optimised for SAIC terminals, especially when the mobiles are able to signal their SAIC/ARP capability to the network.
Capacity gain
50% 40%
ACKNOWLEDGEMENTS
30%
The authors would like to thank Dr. Markku Pukkila and Mr. Luigi Mattellini, both from Nokia Research Center, for providing the link level simulation inputs.
20% 10% 0% 95%
98%
REFERENCES
Network QoS
Figure 6. SAIC capacity gains with 100% SAIC penetration using two different antenna patterns and four different performance criteria. Figure 7 show the gains for all the cases, including partial SAIC. It can be seen that rather significant network gains can be achieved already with moderate SAIC proportions. For instance, 15-20% gains can be achieved when 40% of the terminals have a SAIC receiver.
[1]
“Work Item Description. Title: Single Antenna Receiver Interference Cancellation (SAIC)”. GP-023400, 3GPP TSG GERAN#12 Sophia-Antipolis, France, 18-22 November 2002.
[2]
Ranta, P.A.; Hottinen, A.; Honkasalo, Z.-C. “Co-channel interference cancelling receiver for TDMA mobile systems” in Proceedings of IEEE International Conference on Communications (ICC), Seattle, USA, 1995.
[3]
H. Trigui and D. Slock, ‘Cochannel Interference Cancellation Within the Current GSM Standard’, Proc. of the IEEE 1998 International Conference on Universal Personal Communications, October 5-9, 1998, Florence, Italy.
[4]
Universal Mobile Telecommunications System (UMTS): Selection procedures for the choice of radio transmission technologies of the UMTS. UMTS 30.03 version 3.2.0. ETSI Technical Report 101 112 (199804).
[5]
Halonen T., Melero J., Romero J., GSM, GPRS & EDGE Performance: Evolution towards 3G UMTS, John Wiley & Sons, 2002.
[6]
GP-040409 "The effect of SAIC on GPRS performance ". 3GPP TSG GERAN #18: Reykjavik, Iceland. 2-6 February 2004. Source: Nokia.
[7]
GP-032648 "The effect of SAIC terminal penetration on non-SAIC terminal performance". 3GPP TSG GERAN #17: Budapest, Hungary, 17-21 Nov. 2003. Source: Nokia.
60%
Capacity gain
50% 40% 30% 20% 10% 0% 95% 20% SAIC (A) 60% SAIC (B)
20% SAIC (B) 80% SAIC (A)
98%
Network QoS 40% SAIC (A) 80% SAIC (B)
40% SAIC (B) 100% SAIC (A)
60% SAIC (A) 100% SAIC (B)
Figure 7. Capacity gains with 65-degree antenna and partial SAIC penetration.
VI. CONCLUSIONS In this paper, the effect of MS SAIC receiver is evaluated on the network level. Comprehensive performance simulations show that SAIC is a very promising capacity feature for